X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/2a3f4b595b2fe9181ce2d338f0234f84b7aff58d..8ba5ebec745c34478bd1082a07edcb76aab3aefb:/src/simix/smx_network.c diff --git a/src/simix/smx_network.c b/src/simix/smx_network.c index 6b4922ff27..0ee16e8772 100644 --- a/src/simix/smx_network.c +++ b/src/simix/smx_network.c @@ -18,16 +18,14 @@ unsigned long int smx_total_comms = 0; static void SIMIX_waitany_req_remove_from_actions(smx_req_t req); static void SIMIX_comm_copy_data(smx_action_t comm); static smx_action_t SIMIX_comm_new(e_smx_comm_type_t type); -static void SIMIX_comm_remove_from_processes(smx_action_t action); static XBT_INLINE void SIMIX_rdv_push(smx_rdv_t rdv, smx_action_t comm); -static XBT_INLINE void SIMIX_rdv_remove(smx_rdv_t rdv, smx_action_t comm); static smx_action_t SIMIX_rdv_get_request(smx_rdv_t rdv, e_smx_comm_type_t type, int (*match_fun)(void *, void *), void *); static void SIMIX_rdv_free(void *data); void SIMIX_network_init(void) { - rdv_points = xbt_dict_new(); + rdv_points = xbt_dict_new_homogeneous(SIMIX_rdv_free); } void SIMIX_network_exit(void) @@ -50,7 +48,7 @@ smx_rdv_t SIMIX_rdv_create(const char *name) rdv->comm_fifo = xbt_fifo_new(); if (rdv->name) - xbt_dict_set(rdv_points, rdv->name, rdv, SIMIX_rdv_free); + xbt_dict_set(rdv_points, rdv->name, rdv, NULL); } return rdv; } @@ -109,7 +107,7 @@ static XBT_INLINE void SIMIX_rdv_push(smx_rdv_t rdv, smx_action_t comm) * \param rdv The rendez-vous point * \param comm The communication request */ -static XBT_INLINE void SIMIX_rdv_remove(smx_rdv_t rdv, smx_action_t comm) +XBT_INLINE void SIMIX_rdv_remove(smx_rdv_t rdv, smx_action_t comm) { xbt_fifo_remove(rdv->comm_fifo, comm); comm->comm.rdv = NULL; @@ -135,7 +133,7 @@ smx_action_t SIMIX_rdv_get_request(smx_rdv_t rdv, e_smx_comm_type_t type, xbt_fifo_item_t item; void* req_data = NULL; - xbt_fifo_foreach(rdv->comm_fifo, item, action, smx_action_t){ + xbt_fifo_foreach(rdv->comm_fifo, item, action, smx_action_t) { if (action->comm.type == SIMIX_COMM_SEND) { req_data = action->comm.src_data; } else if (action->comm.type == SIMIX_COMM_RECEIVE) { @@ -188,7 +186,7 @@ int SIMIX_comm_has_recv_match(smx_rdv_t rdv, int (*match_fun)(void*, void*), voi smx_action_t action; xbt_fifo_item_t item; - xbt_fifo_foreach(rdv->comm_fifo, item, action, smx_action_t){ + xbt_fifo_foreach(rdv->comm_fifo, item, action, smx_action_t) { if (action->comm.type == SIMIX_COMM_RECEIVE && (!match_fun || match_fun(data, action->comm.dst_data))) { XBT_DEBUG("Found a matching communication action %p", action); @@ -214,6 +212,7 @@ smx_action_t SIMIX_comm_new(e_smx_comm_type_t type) /* alloc structures */ act = xbt_mallocator_get(simix_global->action_mallocator); + act->type = SIMIX_ACTION_COMMUNICATE; act->state = SIMIX_WAITING; @@ -242,12 +241,14 @@ smx_action_t SIMIX_comm_new(e_smx_comm_type_t type) */ void SIMIX_comm_destroy(smx_action_t action) { - XBT_DEBUG("Destroy action %p (refcount:%d)", action, action->comm.refcount); + XBT_DEBUG("Destroy action %p (refcount: %d), state: %d", + action, action->comm.refcount, action->state); - if (action->comm.refcount <= 0) + if (action->comm.refcount <= 0) { + xbt_backtrace_display_current(); xbt_die("the refcount of comm %p is already 0 before decreasing it. " "That's a bug!", action); - + } action->comm.refcount--; if (action->comm.refcount > 0) return; @@ -264,7 +265,8 @@ void SIMIX_comm_destroy(smx_action_t action) if (action->comm.detached && action->state != SIMIX_DONE) { /* the communication has failed and was detached: * we have to free the buffer */ - ((void_f_pvoid_t) action->comm.src_data)(action->comm.src_buff); + action->comm.clean_fun(action->comm.src_buff); + action->comm.src_buff = NULL; } xbt_mallocator_release(simix_global->action_mallocator, action); @@ -294,7 +296,9 @@ void SIMIX_comm_destroy_internal_actions(smx_action_t action) smx_action_t SIMIX_comm_isend(smx_process_t src_proc, smx_rdv_t rdv, double task_size, double rate, void *src_buff, size_t src_buff_size, - int (*match_fun)(void *, void *), void *data, + int (*match_fun)(void *, void *), + void (*clean_fun)(void *), // used to free the action in case of problem after a detached send + void *data, int detached) { smx_action_t action; @@ -317,6 +321,9 @@ smx_action_t SIMIX_comm_isend(smx_process_t src_proc, smx_rdv_t rdv, if (detached) { action->comm.detached = 1; action->comm.refcount--; + action->comm.clean_fun = clean_fun; + } else { + action->comm.clean_fun = NULL; } /* Setup the communication request */ @@ -333,7 +340,7 @@ smx_action_t SIMIX_comm_isend(smx_process_t src_proc, smx_rdv_t rdv, } SIMIX_comm_start(action); - return action; + return (detached ? NULL : action); } smx_action_t SIMIX_comm_irecv(smx_process_t dst_proc, smx_rdv_t rdv, @@ -373,6 +380,7 @@ smx_action_t SIMIX_comm_irecv(smx_process_t dst_proc, smx_rdv_t rdv, void SIMIX_pre_comm_wait(smx_req_t req, smx_action_t action, double timeout, int idx) { + /* the request may be a wait, a send or a recv */ surf_action_t sleep; @@ -486,8 +494,10 @@ void SIMIX_pre_comm_waitany(smx_req_t req, int idx) } xbt_dynar_foreach(actions, cursor, action){ - /* Associate this request to the action */ + /* associate this request to the the action */ xbt_fifo_push(action->request_list, req); + + /* see if the action is already finished */ if (action->state != SIMIX_WAITING && action->state != SIMIX_RUNNING){ SIMIX_comm_finish(action); break; @@ -547,13 +557,15 @@ XBT_INLINE void SIMIX_comm_start(smx_action_t action) } } +/** + * \brief Answers the SIMIX requests associated to a communication action. + * \param action a finished communication action + */ void SIMIX_comm_finish(smx_action_t action) { - unsigned int destroy_count = 0; + volatile unsigned int destroy_count = 0; smx_req_t req; - SIMIX_comm_remove_from_processes(action); - while ((req = xbt_fifo_shift(action->request_list))) { /* If a waitany request is waiting for this action to finish, then remove @@ -624,11 +636,18 @@ void SIMIX_comm_finish(smx_action_t action) case SIMIX_LINK_FAILURE: TRY { - XBT_DEBUG("Link failure in action %p between '%s' and '%s': posting an exception to the issuer: %s (%p)", - action, - action->comm.src_proc ? action->comm.src_proc->smx_host->name : NULL, - action->comm.dst_proc ? action->comm.dst_proc->smx_host->name : NULL, - req->issuer->name, req->issuer); + XBT_DEBUG("Link failure in action %p between '%s' and '%s': posting an exception to the issuer: %s (%p) detached:%d", + action, + action->comm.src_proc ? action->comm.src_proc->smx_host->name : NULL, + action->comm.dst_proc ? action->comm.dst_proc->smx_host->name : NULL, + req->issuer->name, req->issuer, action->comm.detached); + if (action->comm.src_proc == req->issuer) { + XBT_DEBUG("I'm source"); + } else if (action->comm.dst_proc == req->issuer) { + XBT_DEBUG("I'm dest"); + } else { + XBT_DEBUG("I'm neither source nor dest"); + } THROWF(network_error, 0, "Link failure"); } CATCH(req->issuer->running_ctx->exception) { @@ -636,8 +655,22 @@ void SIMIX_comm_finish(smx_action_t action) } break; + case SIMIX_CANCELED: + TRY { + if (req->issuer == action->comm.dst_proc) { + THROWF(cancel_error, 0, "Communication canceled by the sender"); + } + else { + THROWF(cancel_error, 0, "Communication canceled by the receiver"); + } + } + CATCH(req->issuer->running_ctx->exception) { + req->issuer->doexception = 1; + } + break; + default: - THROW_IMPOSSIBLE; + xbt_die("Unexpected action state in SIMIX_comm_finish: %d", action->state); } /* if there is an exception during a waitany or a testany, indicate the position of the failed communication */ @@ -651,6 +684,7 @@ void SIMIX_comm_finish(smx_action_t action) } req->issuer->waiting_action = NULL; + xbt_fifo_remove(req->issuer->comms, action); SIMIX_request_answer(req); destroy_count++; } @@ -659,6 +693,10 @@ void SIMIX_comm_finish(smx_action_t action) SIMIX_comm_destroy(action); } +/** + * \brief This function is called when a Surf communication action is finished. + * \param action the corresponding Simix communication + */ void SIMIX_post_comm(smx_action_t action) { /* Update action state */ @@ -675,9 +713,10 @@ void SIMIX_post_comm(smx_action_t action) surf_workstation_model->action_state_get(action->comm.dst_timeout) == SURF_ACTION_FAILED) action->state = SIMIX_DST_HOST_FAILURE; else if (action->comm.surf_comm && - surf_workstation_model->action_state_get(action->comm.surf_comm) == SURF_ACTION_FAILED) + surf_workstation_model->action_state_get(action->comm.surf_comm) == SURF_ACTION_FAILED) { + XBT_DEBUG("Puta madre. Surf says that the link broke"); action->state = SIMIX_LINK_FAILURE; - else + } else action->state = SIMIX_DONE; XBT_DEBUG("SIMIX_post_comm: comm %p, state %d, src_proc %p, dst_proc %p, detached: %d", @@ -686,28 +725,19 @@ void SIMIX_post_comm(smx_action_t action) /* destroy the surf actions associated with the Simix communication */ SIMIX_comm_destroy_internal_actions(action); - /* if there are requests associated with the action, then answer them */ - if (xbt_fifo_size(action->request_list)) { - SIMIX_comm_finish(action); - } - else { - SIMIX_comm_remove_from_processes(action); - } -} - -/** - * \brief Removes a communication action from the list of pending communications - * of both processes (if they still exist) - * \param action a communication action - */ -static void SIMIX_comm_remove_from_processes(smx_action_t action) { - + /* remove the communication action from the list of pending communications + * of both processes (if they still exist) */ if (action->comm.src_proc) { xbt_fifo_remove(action->comm.src_proc->comms, action); } if (action->comm.dst_proc) { xbt_fifo_remove(action->comm.dst_proc->comms, action); } + + /* if there are requests associated with the action, then answer them */ + if (xbt_fifo_size(action->request_list)) { + SIMIX_comm_finish(action); + } } void SIMIX_comm_cancel(smx_action_t action) @@ -716,12 +746,11 @@ void SIMIX_comm_cancel(smx_action_t action) /* so remove from it and delete it */ if (action->state == SIMIX_WAITING) { SIMIX_rdv_remove(action->comm.rdv, action); - action->state = SIMIX_FAILED; + action->state = SIMIX_CANCELED; } - else if (!MC_IS_ENABLED + else if (!MC_IS_ENABLED /* when running the MC there are no surf actions */ && (action->state == SIMIX_READY || action->state == SIMIX_RUNNING)) { - /* when running the MC there are no surf actions */ surf_workstation_model->action_cancel(action->comm.surf_comm); } } @@ -828,25 +857,36 @@ XBT_INLINE int SIMIX_comm_is_latency_bounded(smx_action_t action) /******************************************************************************/ /* SIMIX_comm_copy_data callbacks */ /******************************************************************************/ -static void (*SIMIX_comm_copy_data_callback) (smx_action_t, size_t) = +static void (*SIMIX_comm_copy_data_callback) (smx_action_t, void*, size_t) = &SIMIX_comm_copy_pointer_callback; void -SIMIX_comm_set_copy_data_callback(void (*callback) (smx_action_t, size_t)) +SIMIX_comm_set_copy_data_callback(void (*callback) (smx_action_t, void*, size_t)) { SIMIX_comm_copy_data_callback = callback; } -void SIMIX_comm_copy_pointer_callback(smx_action_t comm, size_t buff_size) +void SIMIX_comm_copy_pointer_callback(smx_action_t comm, void* buff, size_t buff_size) { xbt_assert((buff_size == sizeof(void *)), "Cannot copy %zu bytes: must be sizeof(void*)", buff_size); - *(void **) (comm->comm.dst_buff) = comm->comm.src_buff; + *(void **) (comm->comm.dst_buff) = buff; +} + +void SIMIX_comm_copy_buffer_callback(smx_action_t comm, void* buff, size_t buff_size) +{ + XBT_DEBUG("Copy the data over"); + memcpy(comm->comm.dst_buff, buff, buff_size); } -void SIMIX_comm_copy_buffer_callback(smx_action_t comm, size_t buff_size) +void smpi_comm_copy_data_callback(smx_action_t comm, void* buff, size_t buff_size) { - memcpy(comm->comm.dst_buff, comm->comm.src_buff, buff_size); + XBT_DEBUG("Copy the data over"); + memcpy(comm->comm.dst_buff, buff, buff_size); + if (comm->comm.detached) { // if this is a detached send, the source buffer was duplicated by SMPI sender to make the original buffer available to the application ASAP + xbt_free(buff); + comm->comm.src_buff = NULL; + } } /** @@ -862,8 +902,10 @@ void SIMIX_comm_copy_data(smx_action_t comm) XBT_DEBUG("Copying comm %p data from %s (%p) -> %s (%p) (%zu bytes)", comm, - comm->comm.src_proc->smx_host->name, comm->comm.src_buff, - comm->comm.dst_proc->smx_host->name, comm->comm.dst_buff, buff_size); + comm->comm.src_proc ? comm->comm.src_proc->smx_host->name : "a finished process", + comm->comm.src_buff, + comm->comm.dst_proc ? comm->comm.dst_proc->smx_host->name : "a finished process", + comm->comm.dst_buff, buff_size); /* Copy at most dst_buff_size bytes of the message to receiver's buffer */ if (comm->comm.dst_buff_size) @@ -873,10 +915,8 @@ void SIMIX_comm_copy_data(smx_action_t comm) if (comm->comm.dst_buff_size) *comm->comm.dst_buff_size = buff_size; - if (buff_size == 0) - return; - - (*SIMIX_comm_copy_data_callback) (comm, buff_size); + if (buff_size > 0) + SIMIX_comm_copy_data_callback (comm, comm->comm.src_buff, buff_size); /* Set the copied flag so we copy data only once */ /* (this function might be called from both communication ends) */